gum::BayesNetInference< GUM_SCALAR > Class Template Reference

<agrum/BN/inference/BayesNetInference.h> More...

#include <BayesNetInference.h>

Inheritance diagram for gum::BayesNetInference< GUM_SCALAR >:

Collaboration diagram for gum::BayesNetInference< GUM_SCALAR >:

Public Member Functions
Constructors / Destructors
current state of the inference BayesNetInference can be in one of 4 different states: OutdatedStructure: in this state, the inference is fully unprepared to be applied because some events changed the "logical" structure of the BN: for instance a node received a hard evidence, which implies that its outgoing arcs can be removed from the BN, hence involving a structural change in the BN. As a consequence, the (incremental) inference (probably) needs a significant amount of preparation to be ready for the next inference. In a Lazy propagation, for instance, this step amounts to compute a new join tree, hence a new structure in which inference will be applied. Note that classes that inherit from BayesNetInference may be smarter than BayesNetInference and may, in some situations, find out that their data structures are still ok for inference and, therefore, only resort to perform the actions related to the OutdatedPotentials state. As an example, consider a LazyPropagation inference in Bayes Net A->B->C->D->E in which C has received hard evidence e_C and E is the only target. In this case, A and B are not needed for inference, the only potentials that matter are P(D|e_C) and P(E|D). So the smallest join tree needed for inference contains only one clique DE. Now, adding new evidence e_A on A has no impact on E given hard evidence e_C. In this case, LazyPropagation can be smart and not update its join tree. OutdatedPotentials: in this state, the structure of the BN remains unchanged, only some potentials stored in it have changed. Therefore, the inference probably just needs to invalidate some already computed potentials to be ready. Only a light amount of preparation is needed to be able to perform inference. Ready4Inference: in this state, all the data structures are ready for inference. There just remains to perform the inference computations. Done: the heavy computations of inference have been done. There might still remain a few light computations to perform to get the posterior potentials we need. Typically, in Lazy Propagation, all the messages in the join tree have been computed but, to get the potentials, we still need to perform the combinations of the potentials in the cliques with the messages sent to the cliques. In some inference algorithms, this step may even be empty.
	BayesNetInference (const IBayesNet< GUM_SCALAR > *bn)
	default constructor More...
	BayesNetInference ()
	default constructor with a null BN (useful for virtual inheritance) More...
virtual	~BayesNetInference ()
	destructor More...
Accessors / Modifiers
virtual const GraphicalModel &	model () const final
	Returns a constant reference over the IBayesNet referenced by this class. More...
virtual const NodeProperty< Size > &	domainSizes () const final
	get the domain sizes of the random variables of the model More...
virtual bool	isInferenceReady () const noexcept final
	returns whether the inference object is in a ready state More...
virtual bool	isInferenceOutdatedStructure () const noexcept final
	returns whether the inference object is in a OutdatedStructure state More...
virtual bool	isInferenceOutdatedPotentials () const noexcept final
	returns whether the inference object is in a OutdatedPotential state More...
virtual bool	isInferenceDone () const noexcept final
	returns whether the inference object is in a InferenceDone state More...
virtual void	prepareInference () final
	prepare the internal inference structures for the next inference More...
virtual void	makeInference () final
	perform the heavy computations needed to compute the targets' posteriors More...
virtual void	clear ()
	clears all the data structures allocated for the last inference More...
virtual StateOfInference	state () const noexcept final
	returns the state of the inference engine More...
Evidence
virtual void	addEvidence (NodeId id, const Idx val) final
	adds a new hard evidence on node id More...
virtual void	addEvidence (const std::string &nodeName, const Idx val) final
	adds a new hard evidence on node named nodeName More...
virtual void	addEvidence (NodeId id, const std::string &label) final
	adds a new hard evidence on node id More...
virtual void	addEvidence (const std::string &nodeName, const std::string &label) final
	adds a new hard evidence on node named nodeName More...
virtual void	addEvidence (NodeId id, const std::vector< GUM_SCALAR > &vals) final
	adds a new evidence on node id (might be soft or hard) More...
virtual void	addEvidence (const std::string &nodeName, const std::vector< GUM_SCALAR > &vals) final
	adds a new evidence on node named nodeName (might be soft or hard) More...
virtual void	addEvidence (const Potential< GUM_SCALAR > &pot) final
	adds a new evidence on node id (might be soft or hard) More...
virtual void	addEvidence (Potential< GUM_SCALAR > &&pot) final
	adds a new evidence on node id (might be soft or hard) More...
virtual void	addSetOfEvidence (const Set< const Potential< GUM_SCALAR > * > &potset) final
	adds a new set of evidence More...
virtual void	addListOfEvidence (const List< const Potential< GUM_SCALAR > * > &potlist) final
	adds a new list of evidence More...
virtual void	chgEvidence (NodeId id, const Idx val) final
	change the value of an already existing hard evidence More...
virtual void	chgEvidence (const std::string &nodeName, const Idx val) final
	change the value of an already existing hard evidence More...
virtual void	chgEvidence (NodeId id, const std::string &label) final
	change the value of an already existing hard evidence More...
virtual void	chgEvidence (const std::string &nodeName, const std::string &label) final
	change the value of an already existing hard evidence More...
virtual void	chgEvidence (NodeId id, const std::vector< GUM_SCALAR > &vals) final
	change the value of an already existing evidence (might be soft or hard) More...
virtual void	chgEvidence (const std::string &nodeName, const std::vector< GUM_SCALAR > &vals) final
	change the value of an already existing evidence (might be soft or hard) More...
virtual void	chgEvidence (const Potential< GUM_SCALAR > &pot) final
	change the value of an already existing evidence (might be soft or hard) More...
virtual void	eraseAllEvidence () final
	removes all the evidence entered into the network More...
virtual void	eraseEvidence (NodeId id) final
	removed the evidence, if any, corresponding to node id More...
virtual void	eraseEvidence (const std::string &nodeName) final
	removed the evidence, if any, corresponding to node of name nodeName More...
virtual bool	hasEvidence () const final
	indicates whether some node(s) have received evidence More...
virtual bool	hasEvidence (NodeId id) const final
	indicates whether node id has received an evidence More...
virtual bool	hasEvidence (const std::string &nodeName) const final
	indicates whether node id has received an evidence More...
virtual bool	hasHardEvidence (NodeId id) const final
	indicates whether node id has received a hard evidence More...
virtual bool	hasHardEvidence (const std::string &nodeName) const final
	indicates whether node id has received a hard evidence More...
virtual bool	hasSoftEvidence (NodeId id) const final
	indicates whether node id has received a soft evidence More...
virtual bool	hasSoftEvidence (const std::string &nodeName) const final
	indicates whether node id has received a soft evidence More...
virtual Size	nbrEvidence () const final
	returns the number of evidence entered into the Bayesian network More...
virtual Size	nbrHardEvidence () const final
	returns the number of hard evidence entered into the Bayesian network More...
virtual Size	nbrSoftEvidence () const final
	returns the number of soft evidence entered into the Bayesian network More...
const NodeProperty< const Potential< GUM_SCALAR > *> &	evidence () const
	returns the set of evidence More...
const NodeSet &	softEvidenceNodes () const
	returns the set of nodes with soft evidence More...
const NodeSet &	hardEvidenceNodes () const
	returns the set of nodes with hard evidence More...
const NodeProperty< Idx > &	hardEvidence () const
	indicate for each node with hard evidence which value it took More...

Public Types
enum	StateOfInference { StateOfInference::OutdatedStructure, StateOfInference::OutdatedPotentials, StateOfInference::ReadyForInference, StateOfInference::Done }
	current state of the inference More...

Protected Member Functions
virtual void	onStateChanged_ ()=0
	fired when the stage is changed More...
virtual void	onEvidenceAdded_ (const NodeId id, bool isHardEvidence)=0
	fired after a new evidence is inserted More...
virtual void	onEvidenceErased_ (const NodeId id, bool isHardEvidence)=0
	fired before an evidence is removed More...
virtual void	onAllEvidenceErased_ (bool contains_hard_evidence)=0
	fired before all the evidence are erased More...
virtual void	onEvidenceChanged_ (const NodeId id, bool hasChangedSoftHard)=0
	fired after an evidence is changed, in particular when its status (soft/hard) changes More...
virtual void	onModelChanged_ (const GraphicalModel *model)=0
	fired after a new Bayes net has been assigned to the engine More...
virtual void	updateOutdatedStructure_ ()=0
	prepares inference when the latter is in OutdatedStructure state More...
virtual void	updateOutdatedPotentials_ ()=0
	prepares inference when the latter is in OutdatedPotentials state More...
virtual void	makeInference_ ()=0
	called when the inference has to be performed effectively More...
void	setOutdatedStructureState_ ()
	put the inference into an outdated model structure state More...
void	setOutdatedPotentialsState_ ()
	puts the inference into an OutdatedPotentials state if it is not already in an OutdatedStructure state More...
virtual void	setState_ (const StateOfInference state) final
	set the state of the inference engine and call the notification onStateChanged_ when necessary (i.e. when the state has effectively changed). More...
void	setModel_ (const GraphicalModel *model)
void	setModelDuringConstruction_ (const GraphicalModel *model)
	assigns a model during the inference engine construction More...
bool	hasNoModel_ () const

Accessors / Modifiers
friend	MarginalTargetedInference< GUM_SCALAR >
	allow JointInference to access the single targets and inference states More...
friend	JointTargetedInference< GUM_SCALAR >
	allow JointInference to access the single targets and inference states More...
friend	EvidenceInference< GUM_SCALAR >
	allow JointInference to access the single targets and inference states More...
virtual void	setBN (const IBayesNet< GUM_SCALAR > *bn)
	assigns a new BN to the inference engine More...
virtual const IBayesNet< GUM_SCALAR > &	BN () const final
	Returns a constant reference over the IBayesNet referenced by this class. More...
void	_setBayesNetDuringConstruction_ (const IBayesNet< GUM_SCALAR > *bn)
	assigns a BN during the inference engine construction More...

Detailed Description

template<typename GUM_SCALAR>
class gum::BayesNetInference< GUM_SCALAR >

<agrum/BN/inference/BayesNetInference.h>

A generic class for Bayes net inference: handles evidence and the current state of the (incremental) inference

The goal of the BayesNetInference class is twofold: i) handling the common resources of BN inference (bn, soft/hard evidence); ii) propose a general high-level scheme for all the inference methods.

A specialized inference just has to specify how to prepare inference, how to make inference and how to get the posteriors for nodes and set of nodes. The scheme for every inference derived from BayesNetInference will be the same:

1- ie=SpecificInference(bn); // state <- OutdatedStructure 2- set targets and evidence in ie 3- ie.prepareInference(); // state <- Ready4Inference 4.a- change values of evidence in ie // state <- OutdatedPotentials 4.b- change some hard evidence or targets // state <- OutdatedStructure 5- ie.makeInference(); // state <- Done 6- get posteriors 7- goto 2 or 4

BayesNetInference can be in one of 4 different states:

• OutdatedStructure: in this state, the inference is fully unprepared to be applied because some events changed the "logical" structure of the BN: for instance a node received a hard evidence, which implies that its outgoing arcs can be removed from the BN, hence involving a structural change in the BN. As a consequence, the (incremental) inference (probably) needs a significant amount of preparation to be ready for the next inference. In a Lazy propagation, for instance, this step amounts to compute a new join tree, hence a new structure in which inference will be applied. Note that classes that inherit from BayesNetInference may be smarter than BayesNetInference and may, in some situations, find out that their data structures are still ok for inference and, therefore, only resort to perform the actions related to the OutdatedPotentials state. As an example, consider a LazyPropagation inference in Bayes Net A->B->C->D->E in which C has received hard evidence e_C and E is the only target. In this case, A and B are not needed for inference, the only potentials that matter are P(D|e_C) and P(E|D). So the smallest join tree needed for inference contains only one clique DE. Now, adding new evidence e_A on A has no impact on E given hard evidence e_C. In this case, LazyPropagation can be smart and not update its join tree.
• OutdatedPotentials: in this state, the structure of the BN remains unchanged, only some potentials stored in it have changed. Therefore, the inference probably just needs to invalidate some already computed potentials to be ready. Only a light amount of preparation is needed to be able to perform inference.
• Ready4Inference: in this state, all the data structures are ready for inference. There just remains to perform the inference computations.
• Done: the heavy computations of inference have been done. There might still remain a few light computations to perform to get the posterior potentials we need. Typically, in Lazy Propagation, all the messages in the join tree have been computed but, to get the potentials, we still need to perform the combinations of the potentials in the cliques with the messages sent to the cliques. In some inference algorithms, this step may even be empty.

Definition at line 143 of file BayesNetInference.h.

Member Enumeration Documentation

StateOfInference

template<typename GUM_SCALAR >

enum gum::GraphicalModelInference::StateOfInference

stronginherited

current state of the inference

graphicalModelInference can be in one of 4 different states:

• OutdatedStructure: in this state, the inference is fully unprepared to be applied because some events changed the "logical" structure of the model: for instance a node received a hard evidence, which implies that its outgoing arcs can be removed from the model, hence involving a structural change in the model.
• OutdatedPotentials: in this state, the structure of the model remains unchanged, only some potentials stored in it have changed. Therefore, the inference probably just needs to invalidate some already computed potentials to be ready. Only a light amount of preparation is needed to be able to perform inference.
• Ready4Inference: in this state, all the data structures are ready for inference. There just remains to perform the inference computations.
• Done: the heavy computations of inference have been done. There might still remain a few light computations to perform to get the posterior potentials we need.

Enumerator
OutdatedStructure
OutdatedPotentials
ReadyForInference
Done

Definition at line 106 of file graphicalModelInference.h.

    {
      OutdatedStructure,
      OutdatedPotentials,
      ReadyForInference,
      Done
    };

Constructor & Destructor Documentation

BayesNetInference() [1/2]

template<typename GUM_SCALAR >

gum::BayesNetInference< GUM_SCALAR >::BayesNetInference ( const IBayesNet< GUM_SCALAR > *  bn )

explicit

default constructor

Warning

note that, by aGrUM's rule, the BN is not copied but only referenced by the inference algorithm.

Definition at line 35 of file BayesNetInference_tpl.h.

                                                                                      :
      GraphicalModelInference< GUM_SCALAR >(bn) {
    GUM_CONSTRUCTOR(BayesNetInference);
  }

BayesNetInference() [2/2]

template<typename GUM_SCALAR >

gum::BayesNetInference< GUM_SCALAR >::BayesNetInference

(

)

default constructor with a null BN (useful for virtual inheritance)

Warning

BayesNetInference is virtually inherited by MarginalTargetedInference. As a result, the lowest descendant of BayesNetInference will create the latter. To avoid requiring developers to add in the constructors of their inference algorithms a call to BayesNetInference( bn ), we added constructor BayesNetInference(), which will be called automatically by the lowest descendant. Then, MarginalTargetedInference and JointTargetedInference will take care of setting the appropriate bn into BayesNetInference.

Definition at line 43 of file BayesNetInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                     {
    GUM_CONSTRUCTOR(BayesNetInference);
  }

Here is the call graph for this function:

~BayesNetInference()

template<typename GUM_SCALAR >

gum::BayesNetInference< GUM_SCALAR >::~BayesNetInference ( )

virtual

destructor

Definition at line 50 of file BayesNetInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                      {
    GUM_DESTRUCTOR(BayesNetInference);
  }

Here is the call graph for this function:

Member Function Documentation

_setBayesNetDuringConstruction_()

template<typename GUM_SCALAR >

void gum::BayesNetInference< GUM_SCALAR >::_setBayesNetDuringConstruction_ ( const IBayesNet< GUM_SCALAR > *  bn )

private

assigns a BN during the inference engine construction

Definition at line 71 of file BayesNetInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                        {
    this->setModelDuringConstruction_(bn);
  }

Here is the call graph for this function:

addEvidence() [1/8]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::addEvidence ( NodeId  id, const Idx  val  )

finalvirtualinherited

adds a new hard evidence on node id

Exceptions

UndefinedElement	if id does not belong to the Bayesian network
InvalidArgument	if val is not a value for id
InvalidArgument	if id already has an evidence

Definition at line 228 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    addEvidence(_createHardEvidence_(id, val));
  }

Here is the call graph for this function:

addEvidence() [2/8]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::addEvidence ( const std::string &  nodeName, const Idx  val  )

finalvirtualinherited

adds a new hard evidence on node named nodeName

Exceptions

UndefinedElement	if nodeName does not belong to the Bayesian network
InvalidArgument	if val is not a value for id
InvalidArgument	if nodeName already has an evidence

Definition at line 234 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    addEvidence(this->model().idFromName(nodeName), val);
  }

Here is the call graph for this function:

addEvidence() [3/8]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::addEvidence ( NodeId  id, const std::string &  label  )

finalvirtualinherited

adds a new hard evidence on node id

Exceptions

UndefinedElement	if id does not belong to the Bayesian network
InvalidArgument	if val is not a value for id
InvalidArgument	if id already has an evidence

Definition at line 241 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    addEvidence(id, this->model().variable(id)[label]);
  }

Here is the call graph for this function:

addEvidence() [4/8]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::addEvidence ( const std::string &  nodeName, const std::string &  label  )

finalvirtualinherited

adds a new hard evidence on node named nodeName

Exceptions

UndefinedElement	if nodeName does not belong to the Bayesian network
InvalidArgument	if val is not a value for id
InvalidArgument	if nodeName already has an evidence

Definition at line 248 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    NodeId id = this->model().idFromName(nodeName);
    addEvidence(id, this->model().variable(id)[label]);
  }

Here is the call graph for this function:

addEvidence() [5/8]

template<typename GUM_SCALAR >

void gum::GraphicalModelInference< GUM_SCALAR >::addEvidence ( NodeId  id, const std::vector< GUM_SCALAR > &  vals  )

finalvirtualinherited

adds a new evidence on node id (might be soft or hard)

Exceptions

UndefinedElement	if id does not belong to the Bayesian network
InvalidArgument	if id already has an evidence
FatalError	if vals=[0,0,...,0]
InvalidArgument	if the size of vals is different from the domain size of node id

Definition at line 256 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                               {
    // checks that the evidence is meaningful
    if (_model_ == nullptr)
      GUM_ERROR(NullElement,
                "No Bayes net has been assigned to the "
                "inference algorithm");

    if (!_model_->exists(id)) { GUM_ERROR(UndefinedElement, id << " is not a NodeId in the model") }

    if (_model_->variable(id).domainSize() != vals.size()) {
      GUM_ERROR(InvalidArgument,
                "node " << _model_->variable(id)
                        << " and its evidence vector have different sizes.");
    }

    Potential< GUM_SCALAR > pot;
    pot.add(_model_->variable(id));
    pot.fillWith(vals);
    addEvidence(std::move(pot));
  }

Here is the call graph for this function:

addEvidence() [6/8]

template<typename GUM_SCALAR >

void gum::GraphicalModelInference< GUM_SCALAR >::addEvidence ( const std::string &  nodeName, const std::vector< GUM_SCALAR > &  vals  )

finalvirtualinherited

adds a new evidence on node named nodeName (might be soft or hard)

Exceptions

UndefinedElement	if id does not belong to the Bayesian network
InvalidArgument	if nodeName already has an evidence
FatalError	if vals=[0,0,...,0]
InvalidArgument	if the size of vals is different from the domain size of node nodeName

Definition at line 280 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                               {
    addEvidence(this->model().idFromName(nodeName), vals);
  }

Here is the call graph for this function:

addEvidence() [7/8]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::addEvidence ( const Potential< GUM_SCALAR > &  pot )

finalvirtualinherited

adds a new evidence on node id (might be soft or hard)

Exceptions

UndefinedElement	if the potential is defined over several nodes
UndefinedElement	if the node on which the potential is defined does not belong to the Bayesian network
InvalidArgument	if the node of the potential already has an evidence
FatalError	if pot=[0,0,...,0]

Definition at line 325 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                          {
    Potential< GUM_SCALAR > new_pot(pot);
    addEvidence(std::move(new_pot));
  }

Here is the call graph for this function:

addEvidence() [8/8]

template<typename GUM_SCALAR >

void gum::GraphicalModelInference< GUM_SCALAR >::addEvidence ( Potential< GUM_SCALAR > &&  pot )

finalvirtualinherited

adds a new evidence on node id (might be soft or hard)

Exceptions

UndefinedElement	if the potential is defined over several nodes
UndefinedElement	if the node on which the potential is defined does not belong to the Bayesian network
InvalidArgument	if the node of the potential already has an evidence
FatalError	if pot=[0,0,...,0]

Definition at line 287 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                       {
    // check if the potential corresponds to an evidence
    if (pot.nbrDim() != 1) { GUM_ERROR(InvalidArgument, pot << " is not mono-dimensional.") }
    if (_model_ == nullptr)
      GUM_ERROR(NullElement,
                "No Bayes net has been assigned to the "
                "inference algorithm");

    NodeId id = _model_->nodeId(pot.variable(0));

    if (hasEvidence(id)) {
      GUM_ERROR(InvalidArgument,
                " node " << id << " already has an evidence. Please use chgEvidence().");
    }

    // check whether we have a hard evidence (and also check whether the
    // potential only contains 0 (in this case, this will automatically raise
    // an exception) )
    Idx  val;
    bool is_hard_evidence = _isHardEvidence_(pot, val);

    // insert the evidence
    _evidence_.insert(id,
                      new Potential< GUM_SCALAR >(std::forward< Potential< GUM_SCALAR > >(pot)));
    if (is_hard_evidence) {   // pot is deterministic
      _hard_evidence_.insert(id, val);
      _hard_evidence_nodes_.insert(id);
    } else {
      _soft_evidence_nodes_.insert(id);
    }
    setState_(StateOfInference::OutdatedStructure);
    onEvidenceAdded_(id, is_hard_evidence);
  }

Here is the call graph for this function:

addListOfEvidence()

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::addListOfEvidence ( const List< const Potential< GUM_SCALAR > * > &  potlist )

finalvirtualinherited

adds a new list of evidence

Exceptions

UndefinedElement	if some potential is defined over several nodes
UndefinedElement	if the node on which some potential is defined does not belong to the Bayesian network
InvalidArgument	if the node of some potential already has an evidence
FatalError	if pot=[0,0,...,0]

Definition at line 333 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                            {
    for (const auto pot: potlist)
      addEvidence(*pot);
  }

Here is the call graph for this function:

addSetOfEvidence()

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::addSetOfEvidence ( const Set< const Potential< GUM_SCALAR > * > &  potset )

finalvirtualinherited

adds a new set of evidence

Exceptions

UndefinedElement	if some potential is defined over several nodes
UndefinedElement	if the node on which some potential is defined does not belong to the Bayesian network
InvalidArgument	if the node of some potential already has an evidence
FatalError	if pot=[0,0,...,0]

Definition at line 342 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                          {
    for (const auto pot: potset)
      addEvidence(*pot);
  }

Here is the call graph for this function:

BN()

template<typename GUM_SCALAR >

INLINE const IBayesNet< GUM_SCALAR > & gum::BayesNetInference< GUM_SCALAR >::BN ( ) const

finalvirtual

Returns a constant reference over the IBayesNet referenced by this class.

Exceptions

UndefinedElement

is raised if no Bayes net has been assigned to the inference.

Definition at line 57 of file BayesNetInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                  {
    return static_cast< const IBayesNet< GUM_SCALAR >& >(this->model());
  }

Here is the call graph for this function:

chgEvidence() [1/7]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::chgEvidence ( NodeId  id, const Idx  val  )

finalvirtualinherited

change the value of an already existing hard evidence

Exceptions

UndefinedElement	if id does not belong to the Bayesian network
InvalidArgument	if val is not a value for id
InvalidArgument	if id does not already have an evidence

Definition at line 402 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    chgEvidence(_createHardEvidence_(id, val));
  }

Here is the call graph for this function:

chgEvidence() [2/7]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::chgEvidence ( const std::string &  nodeName, const Idx  val  )

finalvirtualinherited

change the value of an already existing hard evidence

Exceptions

UndefinedElement	if nodeName does not belong to the Bayesian network
InvalidArgument	if val is not a value for id
InvalidArgument	if id does not already have an evidence

Definition at line 408 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    chgEvidence(this->model().idFromName(nodeName), val);
  }

Here is the call graph for this function:

chgEvidence() [3/7]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::chgEvidence ( NodeId  id, const std::string &  label  )

finalvirtualinherited

change the value of an already existing hard evidence

Exceptions

UndefinedElement	if id does not belong to the Bayesian network
InvalidArgument	if val is not a value for id
InvalidArgument	if id does not already have an evidence

Definition at line 415 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    chgEvidence(id, this->model().variable(id)[label]);
  }

Here is the call graph for this function:

chgEvidence() [4/7]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::chgEvidence ( const std::string &  nodeName, const std::string &  label  )

finalvirtualinherited

change the value of an already existing hard evidence

Exceptions

UndefinedElement	if nodeName does not belong to the Bayesian network
InvalidArgument	if val is not a value for id
InvalidArgument	if id does not already have an evidence

Definition at line 422 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    NodeId id = this->model().idFromName(nodeName);
    chgEvidence(id, this->model().variable(id)[label]);
  }

Here is the call graph for this function:

chgEvidence() [5/7]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::chgEvidence ( NodeId  id, const std::vector< GUM_SCALAR > &  vals  )

finalvirtualinherited

change the value of an already existing evidence (might be soft or hard)

Exceptions

UndefinedElement	if id does not belong to the Bayesian network
InvalidArgument	if the node does not already have an evidence
FatalError	if vals=[0,0,...,0]
InvalidArgument	if the size of vals is different from the domain size of node id

Definition at line 431 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                             {
    // check whether this corresponds to an evidence
    if (_model_ == nullptr)
      GUM_ERROR(NullElement,
                "No Bayes net has been assigned to the "
                "inference algorithm");

    if (!_model_->exists(id)) { GUM_ERROR(UndefinedElement, id << " is not a NodeId in the model") }

    if (_model_->variable(id).domainSize() != vals.size()) {
      GUM_ERROR(InvalidArgument,
                "node " << _model_->variable(id) << " and its evidence have different sizes.");
    }

    // create the potential corresponding to vals
    Potential< GUM_SCALAR > pot;
    pot.add(_model_->variable(id));
    pot.fillWith(vals);
    chgEvidence(pot);
  }

Here is the call graph for this function:

chgEvidence() [6/7]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::chgEvidence ( const std::string &  nodeName, const std::vector< GUM_SCALAR > &  vals  )

finalvirtualinherited

change the value of an already existing evidence (might be soft or hard)

Exceptions

UndefinedElement	if nodeName does not belong to the Bayesian network
InvalidArgument	if the node does not already have an evidence
FatalError	if vals=[0,0,...,0]
InvalidArgument	if the size of vals is different from the domain size of node id

Definition at line 456 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                             {
    chgEvidence(this->model().idFromName(nodeName), vals);
  }

Here is the call graph for this function:

chgEvidence() [7/7]

template<typename GUM_SCALAR >

void gum::GraphicalModelInference< GUM_SCALAR >::chgEvidence ( const Potential< GUM_SCALAR > &  pot )

finalvirtualinherited

change the value of an already existing evidence (might be soft or hard)

Exceptions

UndefinedElement	if the potential is defined over several nodes
UndefinedElement	if the node on which the potential is defined does not belong to the Bayesian network
InvalidArgument	if the node of the potential does not already have an evidence
FatalError	if pot=[0,0,...,0]

Definition at line 464 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                            {
    // check if the potential corresponds to an evidence
    if (pot.nbrDim() != 1) {
      GUM_ERROR(InvalidArgument, pot << " is not a mono-dimensional potential.")
    }
    if (_model_ == nullptr)
      GUM_ERROR(NullElement,
                "No Bayes net has been assigned to the "
                "inference algorithm");

    NodeId id = _model_->nodeId(pot.variable(0));

    if (!hasEvidence(id)) {
      GUM_ERROR(InvalidArgument, id << " has no evidence. Please use addEvidence().")
    }

    // check whether we have a hard evidence (and also check whether the
    // potential only contains 0 (in this case, this will automatically raise
    // an exception) )
    Idx  val;
    bool is_hard_evidence = _isHardEvidence_(pot, val);

    // modify the evidence already stored
    const Potential< GUM_SCALAR >* localPot = _evidence_[id];
    Instantiation                  I(pot);
    for (I.setFirst(); !I.end(); I.inc()) {
      localPot->set(I, pot[I]);
    }

    // the inference state will be different
    // whether evidence change from Hard to Soft or not.
    bool hasChangedSoftHard = false;

    if (is_hard_evidence) {
      if (!hasHardEvidence(id)) {
        hasChangedSoftHard = true;
        _hard_evidence_.insert(id, val);
        _hard_evidence_nodes_.insert(id);
        _soft_evidence_nodes_.erase(id);
      } else {
        _hard_evidence_[id] = val;
      }
    } else {
      if (hasHardEvidence(id)) {   // evidence was hard
        _hard_evidence_.erase(id);
        _hard_evidence_nodes_.erase(id);
        _soft_evidence_nodes_.insert(id);
        hasChangedSoftHard = true;
      }
    }

    if (hasChangedSoftHard) {
      setState_(StateOfInference::OutdatedStructure);
    } else {
      if (!isInferenceOutdatedStructure()) { setState_(StateOfInference::OutdatedPotentials); }
    }

    onEvidenceChanged_(id, hasChangedSoftHard);
  }

Here is the call graph for this function:

clear()

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::clear ( )

virtualinherited

clears all the data structures allocated for the last inference

Reimplemented in gum::ShaferShenoyLIMIDInference< GUM_SCALAR >.

Definition at line 139 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                           {
    eraseAllEvidence();
    setState_(StateOfInference::OutdatedStructure);
  }

Here is the call graph for this function:

domainSizes()

template<typename GUM_SCALAR >

INLINE const NodeProperty< Size > & gum::GraphicalModelInference< GUM_SCALAR >::domainSizes ( ) const

finalvirtualinherited

get the domain sizes of the random variables of the model

Definition at line 159 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                              {
    return _domain_sizes_;
  }

Here is the call graph for this function:

eraseAllEvidence()

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::eraseAllEvidence ( )

finalvirtualinherited

removes all the evidence entered into the network

Definition at line 553 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                      {
    bool has_hard_evidence = !_hard_evidence_.empty();
    this->onAllEvidenceErased_(has_hard_evidence);

    for (const auto& pair: _evidence_) {
      if (pair.second != nullptr) { delete (pair.second); }
    }

    _evidence_.clear();
    _hard_evidence_.clear();
    _hard_evidence_nodes_.clear();
    _soft_evidence_nodes_.clear();

    if (has_hard_evidence) {
      setState_(StateOfInference::OutdatedStructure);
    } else {
      if (!isInferenceOutdatedStructure()) { setState_(StateOfInference::OutdatedPotentials); }
    }
  }

Here is the call graph for this function:

eraseEvidence() [1/2]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::eraseEvidence ( NodeId  id )

finalvirtualinherited

removed the evidence, if any, corresponding to node id

Definition at line 527 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                            {
    if (hasEvidence(id)) {
      if (hasHardEvidence(id)) {
        onEvidenceErased_(id, true);
        _hard_evidence_.erase(id);
        _hard_evidence_nodes_.erase(id);
        setState_(StateOfInference::OutdatedStructure);
      } else {
        onEvidenceErased_(id, false);
        _soft_evidence_nodes_.erase(id);
        if (!isInferenceOutdatedStructure()) { setState_(StateOfInference::OutdatedPotentials); }
      }

      delete (_evidence_[id]);
      _evidence_.erase(id);
    }
  }

Here is the call graph for this function:

eraseEvidence() [2/2]

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::eraseEvidence ( const std::string &  nodeName )

finalvirtualinherited

removed the evidence, if any, corresponding to node of name nodeName

Definition at line 546 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                            {
    eraseEvidence(this->model().idFromName(nodeName));
  }

Here is the call graph for this function:

evidence()

template<typename GUM_SCALAR >

INLINE const NodeProperty< const Potential< GUM_SCALAR > *> & gum::GraphicalModelInference< GUM_SCALAR >::evidence ( ) const

inherited

returns the set of evidence

Definition at line 605 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                     {
    return _evidence_;
  }

Here is the call graph for this function:

hardEvidence()

template<typename GUM_SCALAR >

INLINE const NodeProperty< Idx > & gum::GraphicalModelInference< GUM_SCALAR >::hardEvidence ( ) const

inherited

indicate for each node with hard evidence which value it took

Definition at line 597 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                              {
    return _hard_evidence_;
  }

Here is the call graph for this function:

hardEvidenceNodes()

template<typename GUM_SCALAR >

INLINE const NodeSet & gum::GraphicalModelInference< GUM_SCALAR >::hardEvidenceNodes ( ) const

inherited

returns the set of nodes with hard evidence

the set of nodes that received hard evidence

Definition at line 619 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                       {
    return _hard_evidence_nodes_;
  }

Here is the call graph for this function:

hasEvidence() [1/3]

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::hasEvidence ( ) const

finalvirtualinherited

indicates whether some node(s) have received evidence

Definition at line 351 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                       {
    return !_evidence_.empty();
  }

Here is the call graph for this function:

hasEvidence() [2/3]

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::hasEvidence ( NodeId  id ) const

finalvirtualinherited

indicates whether node id has received an evidence

Definition at line 358 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                {
    return _evidence_.exists(id);
  }

Here is the call graph for this function:

hasEvidence() [3/3]

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::hasEvidence ( const std::string &  nodeName ) const

finalvirtualinherited

indicates whether node id has received an evidence

Definition at line 380 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                       {
    return hasEvidence(this->model().idFromName(nodeName));
  }

Here is the call graph for this function:

hasHardEvidence() [1/2]

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::hasHardEvidence ( NodeId  id ) const

finalvirtualinherited

indicates whether node id has received a hard evidence

Definition at line 365 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                    {
    return _hard_evidence_nodes_.exists(id);
  }

Here is the call graph for this function:

hasHardEvidence() [2/2]

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::hasHardEvidence ( const std::string &  nodeName ) const

finalvirtualinherited

indicates whether node id has received a hard evidence

Definition at line 388 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                           {
    return hasHardEvidence(this->model().idFromName(nodeName));
  }

Here is the call graph for this function:

hasNoModel_()

template<typename GUM_SCALAR >

bool gum::GraphicalModelInference< GUM_SCALAR >::hasNoModel_ ( ) const

inlineprotectedinherited

Definition at line 527 of file graphicalModelInference.h.

{ return _model_ == nullptr; };

hasSoftEvidence() [1/2]

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::hasSoftEvidence ( NodeId  id ) const

finalvirtualinherited

indicates whether node id has received a soft evidence

Definition at line 372 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                    {
    return _soft_evidence_nodes_.exists(id);
  }

Here is the call graph for this function:

hasSoftEvidence() [2/2]

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::hasSoftEvidence ( const std::string &  nodeName ) const

finalvirtualinherited

indicates whether node id has received a soft evidence

Definition at line 396 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                           {
    return hasSoftEvidence(this->model().idFromName(nodeName));
  }

Here is the call graph for this function:

isInferenceDone()

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::isInferenceDone ( ) const

finalvirtualnoexceptinherited

returns whether the inference object is in a InferenceDone state

The inference object is in a done state when the posteriors can be retrieved without performing a new inference, i.e., all the heavy computations have already been performed. Typically, in a junction tree algorithm, this corresponds to a situation in which all the messages needed in the JT have been computed and sent.

Definition at line 84 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                    {
    return (_state_ == StateOfInference::Done);
  }

Here is the call graph for this function:

isInferenceOutdatedPotentials()

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::isInferenceOutdatedPotentials ( ) const

finalvirtualnoexceptinherited

returns whether the inference object is in a OutdatedPotential state

Definition at line 79 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    return (_state_ == StateOfInference::OutdatedPotentials);
  }

Here is the call graph for this function:

isInferenceOutdatedStructure()

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::isInferenceOutdatedStructure ( ) const

finalvirtualnoexceptinherited

returns whether the inference object is in a OutdatedStructure state

Definition at line 73 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                                 {
    return (_state_ == StateOfInference::OutdatedStructure);
  }

Here is the call graph for this function:

isInferenceReady()

template<typename GUM_SCALAR >

INLINE bool gum::GraphicalModelInference< GUM_SCALAR >::isInferenceReady ( ) const

finalvirtualnoexceptinherited

returns whether the inference object is in a ready state

Definition at line 68 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                     {
    return (_state_ == StateOfInference::ReadyForInference);
  }

Here is the call graph for this function:

makeInference()

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::makeInference ( )

finalvirtualinherited

perform the heavy computations needed to compute the targets' posteriors

In a Junction tree propagation scheme, for instance, the heavy computations are those of the messages sent in the JT. This is precisely what makeInference should compute. Later, the computations of the posteriors can be done "lightly" by multiplying and projecting those messages.

Definition at line 664 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                   {
    if (isInferenceDone()) { return; }

    if (!isInferenceReady()) { prepareInference(); }

    makeInference_();

    setState_(StateOfInference::Done);
  }

Here is the call graph for this function:

makeInference_()

template<typename GUM_SCALAR >

virtual void gum::GraphicalModelInference< GUM_SCALAR >::makeInference_ ( )

protectedpure virtualinherited

called when the inference has to be performed effectively

Once the inference is done, fillPosterior_ can be called.

Implemented in gum::LazyPropagation< GUM_SCALAR >, gum::SamplingInference< GUM_SCALAR >, gum::VariableElimination< GUM_SCALAR >, gum::ShaferShenoyInference< GUM_SCALAR >, gum::ShaferShenoyMNInference< GUM_SCALAR >, gum::ShaferShenoyLIMIDInference< GUM_SCALAR >, and gum::LoopyBeliefPropagation< GUM_SCALAR >.

model()

template<typename GUM_SCALAR >

INLINE const GraphicalModel & gum::GraphicalModelInference< GUM_SCALAR >::model ( ) const

finalvirtualinherited

Returns a constant reference over the IBayesNet referenced by this class.

Exceptions

UndefinedElement

is raised if no Bayes net has been assigned to the inference.

Definition at line 107 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                  {
    if (_model_ == nullptr)
      GUM_ERROR(UndefinedElement,
                "No Bayes net has been assigned to "
                "the inference algorithm.");
    return *_model_;
  }

Here is the call graph for this function:

nbrEvidence()

template<typename GUM_SCALAR >

INLINE Size gum::GraphicalModelInference< GUM_SCALAR >::nbrEvidence ( ) const

finalvirtualinherited

returns the number of evidence entered into the Bayesian network

Definition at line 576 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                       {
    return _evidence_.size();
  }

Here is the call graph for this function:

nbrHardEvidence()

template<typename GUM_SCALAR >

INLINE Size gum::GraphicalModelInference< GUM_SCALAR >::nbrHardEvidence ( ) const

finalvirtualinherited

returns the number of hard evidence entered into the Bayesian network

Definition at line 583 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                           {
    return _hard_evidence_nodes_.size();
  }

Here is the call graph for this function:

nbrSoftEvidence()

template<typename GUM_SCALAR >

INLINE Size gum::GraphicalModelInference< GUM_SCALAR >::nbrSoftEvidence ( ) const

finalvirtualinherited

returns the number of soft evidence entered into the Bayesian network

Definition at line 590 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                           {
    return _soft_evidence_nodes_.size();
  }

Here is the call graph for this function:

onAllEvidenceErased_()

template<typename GUM_SCALAR >

virtual void gum::GraphicalModelInference< GUM_SCALAR >::onAllEvidenceErased_ ( bool  contains_hard_evidence )

protectedpure virtualinherited

fired before all the evidence are erased

Implemented in gum::SamplingInference< GUM_SCALAR >, gum::ShaferShenoyLIMIDInference< GUM_SCALAR >, gum::LazyPropagation< GUM_SCALAR >, gum::VariableElimination< GUM_SCALAR >, gum::ShaferShenoyInference< GUM_SCALAR >, gum::ShaferShenoyMNInference< GUM_SCALAR >, and gum::LoopyBeliefPropagation< GUM_SCALAR >.

onEvidenceAdded_()

template<typename GUM_SCALAR >

virtual void gum::GraphicalModelInference< GUM_SCALAR >::onEvidenceAdded_ ( const NodeId  id, bool  isHardEvidence  )

protectedpure virtualinherited

fired after a new evidence is inserted

Implemented in gum::SamplingInference< GUM_SCALAR >, gum::ShaferShenoyLIMIDInference< GUM_SCALAR >, gum::LazyPropagation< GUM_SCALAR >, gum::VariableElimination< GUM_SCALAR >, gum::ShaferShenoyInference< GUM_SCALAR >, gum::ShaferShenoyMNInference< GUM_SCALAR >, and gum::LoopyBeliefPropagation< GUM_SCALAR >.

onEvidenceChanged_()

template<typename GUM_SCALAR >

virtual void gum::GraphicalModelInference< GUM_SCALAR >::onEvidenceChanged_ ( const NodeId  id, bool  hasChangedSoftHard  )

protectedpure virtualinherited

fired after an evidence is changed, in particular when its status (soft/hard) changes

Parameters

nodeId	the node of the changed evidence
hasChangedSoftHard	true if the evidence has changed from Soft to Hard or from Hard to Soft

Implemented in gum::SamplingInference< GUM_SCALAR >, gum::LazyPropagation< GUM_SCALAR >, gum::ShaferShenoyLIMIDInference< GUM_SCALAR >, gum::VariableElimination< GUM_SCALAR >, gum::ShaferShenoyInference< GUM_SCALAR >, gum::ShaferShenoyMNInference< GUM_SCALAR >, and gum::LoopyBeliefPropagation< GUM_SCALAR >.

onEvidenceErased_()

template<typename GUM_SCALAR >

virtual void gum::GraphicalModelInference< GUM_SCALAR >::onEvidenceErased_ ( const NodeId  id, bool  isHardEvidence  )

protectedpure virtualinherited

fired before an evidence is removed

Implemented in gum::SamplingInference< GUM_SCALAR >, gum::ShaferShenoyLIMIDInference< GUM_SCALAR >, gum::LazyPropagation< GUM_SCALAR >, gum::VariableElimination< GUM_SCALAR >, gum::ShaferShenoyInference< GUM_SCALAR >, gum::ShaferShenoyMNInference< GUM_SCALAR >, and gum::LoopyBeliefPropagation< GUM_SCALAR >.

onModelChanged_()

template<typename GUM_SCALAR >

virtual void gum::GraphicalModelInference< GUM_SCALAR >::onModelChanged_ ( const GraphicalModel *  model )

protectedpure virtualinherited

fired after a new Bayes net has been assigned to the engine

Implemented in gum::JointTargetedMNInference< GUM_SCALAR >, gum::JointTargetedInference< GUM_SCALAR >, gum::SamplingInference< GUM_SCALAR >, gum::MarginalTargetedInference< GUM_SCALAR >, gum::MarginalTargetedMNInference< GUM_SCALAR >, gum::LazyPropagation< GUM_SCALAR >, gum::ShaferShenoyLIMIDInference< GUM_SCALAR >, gum::VariableElimination< GUM_SCALAR >, gum::ShaferShenoyInference< GUM_SCALAR >, and gum::LoopyBeliefPropagation< GUM_SCALAR >.

onStateChanged_()

template<typename GUM_SCALAR >

virtual void gum::GraphicalModelInference< GUM_SCALAR >::onStateChanged_ ( )

protectedpure virtualinherited

fired when the stage is changed

Implemented in gum::SamplingInference< GUM_SCALAR >, gum::LazyPropagation< GUM_SCALAR >, gum::ShaferShenoyLIMIDInference< GUM_SCALAR >, gum::VariableElimination< GUM_SCALAR >, gum::ShaferShenoyInference< GUM_SCALAR >, gum::ShaferShenoyMNInference< GUM_SCALAR >, and gum::LoopyBeliefPropagation< GUM_SCALAR >.

prepareInference()

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::prepareInference ( )

finalvirtualinherited

prepare the internal inference structures for the next inference

Definition at line 645 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                      {
    if (isInferenceReady() || isInferenceDone()) { return; }

    if (_model_ == nullptr)
      GUM_ERROR(NullElement,
                "No model been assigned to the "
                "inference algorithm");

    if (_state_ == StateOfInference::OutdatedStructure)
      updateOutdatedStructure_();
    else
      updateOutdatedPotentials_();

    setState_(StateOfInference::ReadyForInference);
  }

Here is the call graph for this function:

setBN()

template<typename GUM_SCALAR >

void gum::BayesNetInference< GUM_SCALAR >::setBN ( const IBayesNet< GUM_SCALAR > *  bn )

virtual

assigns a new BN to the inference engine

Assigns a new BN to the BayesNetInference engine and sends messages to the descendants of BayesNetInference to inform them that the BN has changed.

Warning

By default, all the nodes of the Bayes net are targets.

note that, by aGrUM's rule, the bn is not copied into the inference engine but only referenced.

Definition at line 64 of file BayesNetInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                               {
    this->setModel_(bn);
  }

Here is the call graph for this function:

setModel_()

template<typename GUM_SCALAR >

void gum::GraphicalModelInference< GUM_SCALAR >::setModel_ ( const GraphicalModel *  model )

protectedinherited

Definition at line 118 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                   {
    clear();
    _model_ = model;
    _computeDomainSizes_();
    onModelChanged_(model);
    setState_(StateOfInference::OutdatedStructure);
  }

Here is the call graph for this function:

setModelDuringConstruction_()

template<typename GUM_SCALAR >

void gum::GraphicalModelInference< GUM_SCALAR >::setModelDuringConstruction_ ( const GraphicalModel *  model )

protectedinherited

assigns a model during the inference engine construction

Definition at line 129 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                  {
    _model_ = model;
    _computeDomainSizes_();
    setState_(StateOfInference::OutdatedStructure);
  }

Here is the call graph for this function:

setOutdatedPotentialsState_()

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::setOutdatedPotentialsState_ ( )

protectedinherited

puts the inference into an OutdatedPotentials state if it is not already in an OutdatedStructure state

OutdatedPotentials: in this state, the structure of the model remains unchanged, only some potentials stored in it have changed. Therefore, the inference probably just needs to invalidate some already computed potentials to be ready. Only a light amount of preparation is needed to be able to perform inference.

Definition at line 638 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                 {
    setState_(StateOfInference::OutdatedPotentials);
  }

Here is the call graph for this function:

setOutdatedStructureState_()

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::setOutdatedStructureState_ ( )

protectedinherited

put the inference into an outdated model structure state

OutdatedStructure: in this state, the inference is fully unprepared to be applied because some events changed the "logical" structure of the model: for instance a node received a hard evidence, which implies that its outgoing arcs can be removed from the model, hence involving a structural change in the model. As a consequence, the (incremental) inference (probably) needs a significant amount of preparation to be ready for the next inference. In a Lazy propagation, for instance, this step amounts to compute a new join tree, hence a new structure in which inference will be applied. Note that classes that inherit from graphicalModelInference may be smarter than graphicalModelInference and may, in some situations, find out that their data structures are still ok for inference and, therefore, only resort to perform the actions related to the OutdatedPotentials state. As an example, consider a LazyPropagation inference in Bayes Net A->B->C->D->E in which C has received hard evidence e_C and E is the only target. In this case, A and B are not needed for inference, the only potentials that matter are P(D|e_C) and P(E|D). So the smallest join tree needed for inference contains only one clique DE. Now, adding new evidence e_A on A has no impact on E given hard evidence e_C. In this case, LazyPropagation can be smart and not update its join tree.

Definition at line 630 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                {
    setState_(StateOfInference::OutdatedStructure);
  }

Here is the call graph for this function:

setState_()

template<typename GUM_SCALAR >

INLINE void gum::GraphicalModelInference< GUM_SCALAR >::setState_ ( const StateOfInference  state )

finalprotectedvirtualinherited

set the state of the inference engine and call the notification onStateChanged_ when necessary (i.e. when the state has effectively changed).

Definition at line 98 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                           {
    if (_state_ != state) {
      _state_ = state;
      onStateChanged_();
    }
  }

Here is the call graph for this function:

softEvidenceNodes()

template<typename GUM_SCALAR >

INLINE const NodeSet & gum::GraphicalModelInference< GUM_SCALAR >::softEvidenceNodes ( ) const

inherited

returns the set of nodes with soft evidence

the set of nodes that received soft evidence

Definition at line 612 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                       {
    return _soft_evidence_nodes_;
  }

Here is the call graph for this function:

state()

template<typename GUM_SCALAR >

INLINE GraphicalModelInference< GUM_SCALAR >::StateOfInference gum::GraphicalModelInference< GUM_SCALAR >::state ( ) const

finalvirtualnoexceptinherited

returns the state of the inference engine

Definition at line 92 of file graphicalModelInference_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                 {
    return _state_;
  }

Here is the call graph for this function:

updateOutdatedPotentials_()

template<typename GUM_SCALAR >

virtual void gum::GraphicalModelInference< GUM_SCALAR >::updateOutdatedPotentials_ ( )

protectedpure virtualinherited

prepares inference when the latter is in OutdatedPotentials state

Note that the values of evidence are not necessarily known and can be changed between updateOutdatedPotentials_ and makeInference_.

Implemented in gum::SamplingInference< GUM_SCALAR >, gum::LazyPropagation< GUM_SCALAR >, gum::VariableElimination< GUM_SCALAR >, gum::ShaferShenoyInference< GUM_SCALAR >, gum::ShaferShenoyMNInference< GUM_SCALAR >, gum::ShaferShenoyLIMIDInference< GUM_SCALAR >, and gum::LoopyBeliefPropagation< GUM_SCALAR >.

updateOutdatedStructure_()

template<typename GUM_SCALAR >

virtual void gum::GraphicalModelInference< GUM_SCALAR >::updateOutdatedStructure_ ( )

protectedpure virtualinherited

prepares inference when the latter is in OutdatedStructure state

Note that the values of evidence are not necessarily known and can be changed between updateOutdatedStructure_ and makeInference_.

Implemented in gum::SamplingInference< GUM_SCALAR >, gum::LazyPropagation< GUM_SCALAR >, gum::VariableElimination< GUM_SCALAR >, gum::ShaferShenoyInference< GUM_SCALAR >, gum::ShaferShenoyMNInference< GUM_SCALAR >, gum::ShaferShenoyLIMIDInference< GUM_SCALAR >, and gum::LoopyBeliefPropagation< GUM_SCALAR >.

Member Data Documentation

EvidenceInference< GUM_SCALAR >

template<typename GUM_SCALAR >

friend gum::BayesNetInference< GUM_SCALAR >::EvidenceInference< GUM_SCALAR >

private

allow JointInference to access the single targets and inference states

Definition at line 239 of file BayesNetInference.h.

JointTargetedInference< GUM_SCALAR >

template<typename GUM_SCALAR >

friend gum::BayesNetInference< GUM_SCALAR >::JointTargetedInference< GUM_SCALAR >

private

allow JointInference to access the single targets and inference states

Definition at line 238 of file BayesNetInference.h.

MarginalTargetedInference< GUM_SCALAR >

template<typename GUM_SCALAR >

friend gum::BayesNetInference< GUM_SCALAR >::MarginalTargetedInference< GUM_SCALAR >

private

allow JointInference to access the single targets and inference states

Definition at line 237 of file BayesNetInference.h.

---

The documentation for this class was generated from the following files:

• agrum/BN/inference/tools/BayesNetInference.h
• agrum/BN/inference/tools/BayesNetInference_tpl.h